GitHub/LineageOS/android_kernel_samsung_universal7580.git
18 years ago[PATCH] x86: cache pollution aware __copy_from_user_ll()
Hiro Yoshioka [Fri, 23 Jun 2006 09:04:16 +0000 (02:04 -0700)]
[PATCH] x86: cache pollution aware __copy_from_user_ll()

Use the x86 cache-bypassing copy instructions for copy_from_user().

Some performance data are

Total of GLOBAL_POWER_EVENTS (CPU cycle samples)

2.6.12.4.orig    1921587
2.6.12.4.nt      1599424
1599424/1921587=83.23% (16.77% reduction)

BSQ_CACHE_REFERENCE (L3 cache miss)
2.6.12.4.orig      57427
2.6.12.4.nt        20858
20858/57427=36.32% (63.7% reduction)

L3 cache miss reduction of __copy_from_user_ll
samples  %
37408    65.1412  vmlinux                  __copy_from_user_ll
23        0.1103  vmlinux                  __copy_user_zeroing_intel_nocache
23/37408=0.061% (99.94% reduction)

Top 5 of 2.6.12.4.nt
Counted GLOBAL_POWER_EVENTS events (time during which processor is not stopped) with a unit mask of 0x01 (mandatory) count 100000
samples  %        app name                 symbol name
128392    8.0274  vmlinux                  __copy_user_zeroing_intel_nocache
64206     4.0143  vmlinux                  journal_add_journal_head
59746     3.7355  vmlinux                  do_get_write_access
47674     2.9807  vmlinux                  journal_put_journal_head
46021     2.8774  vmlinux                  journal_dirty_metadata
pattern9-0-cpu4-0-09011728/summary.out

Counted BSQ_CACHE_REFERENCE events (cache references seen by the bus unit) with a unit mask of 0x3f (multiple flags) count 3000
samples  %        app name                 symbol name
69755     4.2861  vmlinux                  __copy_user_zeroing_intel_nocache
55685     3.4215  vmlinux                  journal_add_journal_head
52371     3.2179  vmlinux                  __find_get_block
45504     2.7960  vmlinux                  journal_put_journal_head
36005     2.2123  vmlinux                  journal_stop
pattern9-0-cpu4-0-09011744/summary.out

Counted BSQ_CACHE_REFERENCE events (cache references seen by the bus unit) with a unit mask of 0x200 (read 3rd level cache miss) count 3000
samples  %        app name                 symbol name
1147      5.4994  vmlinux                  journal_add_journal_head
881       4.2240  vmlinux                  journal_dirty_data
872       4.1809  vmlinux                  blk_rq_map_sg
734       3.5192  vmlinux                  journal_commit_transaction
617       2.9582  vmlinux                  radix_tree_delete
pattern9-0-cpu4-0-09011731/summary.out

iozone results are

original 2.6.12.4 CPU time = 207.768 sec
cache aware       CPU time = 184.783 sec
(three times run)
184.783/207.768=88.94% (11.06% reduction)

original:
pattern9-0-cpu4-0-08191720/iozone.out:  CPU Utilization: Wall time   45.997    CPU time   64.527    CPU utilization 140.28 %
pattern9-0-cpu4-0-08191741/iozone.out:  CPU Utilization: Wall time   46.878    CPU time   71.933    CPU utilization 153.45 %
pattern9-0-cpu4-0-08191743/iozone.out:  CPU Utilization: Wall time   45.152    CPU time   71.308    CPU utilization 157.93 %

cache awre:
pattern9-0-cpu4-0-09011728/iozone.out:  CPU Utilization: Wall time   44.842    CPU time   62.465    CPU utilization 139.30 %
pattern9-0-cpu4-0-09011731/iozone.out:  CPU Utilization: Wall time   44.718    CPU time   59.273    CPU utilization 132.55 %
pattern9-0-cpu4-0-09011744/iozone.out:  CPU Utilization: Wall time   44.367    CPU time   63.045    CPU utilization 142.10 %

Signed-off-by: Hiro Yoshioka <hyoshiok@miraclelinux.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mips: fix number of mremap arguments
Yoichi Yuasa [Fri, 23 Jun 2006 09:04:14 +0000 (02:04 -0700)]
[PATCH] mips: fix number of mremap arguments

mremap syscall takes 5 arguments. Fixed by Ralf Baechle.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Au1550/1200: add missing PSC #define's, make OSS driver use the proper ones
Sergei Shtylyov [Fri, 23 Jun 2006 09:04:13 +0000 (02:04 -0700)]
[PATCH] Au1550/1200: add missing PSC #define's, make OSS driver use the proper ones

Add missing PSC #define's required for the drivers using PSC on DBAu1550
board (also fixing Au1550 PSC3 address) and all Au1200-based boards as
well.  Make the OSS driver use the correct PSC definitions fo each board.

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: clean frv unistd.h
Al Viro [Fri, 23 Jun 2006 09:04:13 +0000 (02:04 -0700)]
[PATCH] frv: clean frv unistd.h

Remove _syscall invocations that aren't used in the kernel.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: trivial cleanups in frv_ksyms.c
Al Viro [Fri, 23 Jun 2006 09:04:12 +0000 (02:04 -0700)]
[PATCH] frv: trivial cleanups in frv_ksyms.c

Remove duplicate EXPORT_SYMBOL annotations from the FRV arch.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: Add missing qualifier to memcpy_fromio() prototype
Al Viro [Fri, 23 Jun 2006 09:04:11 +0000 (02:04 -0700)]
[PATCH] frv: Add missing qualifier to memcpy_fromio() prototype

The memcpy_fromio() function should have a const qualifier on its source
pointer.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: ieee1394 is borken on frv
Al Viro [Fri, 23 Jun 2006 09:04:10 +0000 (02:04 -0700)]
[PATCH] frv: ieee1394 is borken on frv

The ieee1394 assumes it may make direct use of ->count in the semaphore
structure.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Stefan Richter <stefanr@s5r6.in-berlin.de>
Cc: Ben Collins <bcollins@ubuntu.com>
Cc: Jody McIntyre <scjody@modernduck.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: NULL noise removal in frv xchg()
Al Viro [Fri, 23 Jun 2006 09:04:09 +0000 (02:04 -0700)]
[PATCH] frv: NULL noise removal in frv xchg()

Clean up the FRV arch's xchg() function.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: initrd is grossly broken on frv (never built)
Al Viro [Fri, 23 Jun 2006 09:04:09 +0000 (02:04 -0700)]
[PATCH] frv: initrd is grossly broken on frv (never built)

The FRV arch doesn't currently support initrd, so it should be disabled
automatically for the moment.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] ext2 XIP won't build without MMU
Al Viro [Fri, 23 Jun 2006 09:04:08 +0000 (02:04 -0700)]
[PATCH] ext2 XIP won't build without MMU

Disable Ext2 XIP if the kernel is configured in no-MMU mode as the former
won't build.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: wrong syscall
Al Viro [Fri, 23 Jun 2006 09:04:07 +0000 (02:04 -0700)]
[PATCH] frv: wrong syscall

The FRV arch should use fstatat64 not newfstatat.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: misc sparse annotations
Al Viro [Fri, 23 Jun 2006 09:04:07 +0000 (02:04 -0700)]
[PATCH] frv: misc sparse annotations

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: misc __user annotations
Al Viro [Fri, 23 Jun 2006 09:04:06 +0000 (02:04 -0700)]
[PATCH] frv: misc __user annotations

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: binfmt_elf_fdpic __user annotations
Al Viro [Fri, 23 Jun 2006 09:04:05 +0000 (02:04 -0700)]
[PATCH] frv: binfmt_elf_fdpic __user annotations

Add __user annotations to binfmt_elf_fdpic.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: sysctl __user annotations
Al Viro [Fri, 23 Jun 2006 09:04:05 +0000 (02:04 -0700)]
[PATCH] frv: sysctl __user annotations

Add __user annotations to FRV-specific sysctl stuff.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: signal annotations
Al Viro [Fri, 23 Jun 2006 09:04:04 +0000 (02:04 -0700)]
[PATCH] frv: signal annotations

Add annotations to the FRV signal handling for sparse.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: basic __iomem annotations
Al Viro [Fri, 23 Jun 2006 09:04:03 +0000 (02:04 -0700)]
[PATCH] frv: basic __iomem annotations

Add annotations to the FRV I/O handling functions for sparse.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: __user infrastructure
Al Viro [Fri, 23 Jun 2006 09:04:03 +0000 (02:04 -0700)]
[PATCH] frv: __user infrastructure

Add general annotations to the FRV arch for sparse.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SELinux: add security_task_movememory calls to mm code
David Quigley [Fri, 23 Jun 2006 09:04:02 +0000 (02:04 -0700)]
[PATCH] SELinux: add security_task_movememory calls to mm code

This patch inserts security_task_movememory hook calls into memory management
code to enable security modules to mediate this operation between tasks.

Since the last posting, the hook has been renamed following feedback from
Christoph Lameter.

Signed-off-by: David Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
Cc: Andi Kleen <ak@muc.de>
Acked-by: Christoph Lameter <clameter@sgi.com>
Acked-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SELinux: add task_movememory hook
David Quigley [Fri, 23 Jun 2006 09:04:01 +0000 (02:04 -0700)]
[PATCH] SELinux: add task_movememory hook

This patch adds new security hook, task_movememory, to be called when memory
owened by a task is to be moved (e.g.  when migrating pages to a this hook is
identical to the setscheduler implementation, but a separate hook introduced
to allow this check to be specialized in the future if necessary.

Since the last posting, the hook has been renamed following feedback from
Christoph Lameter.

Signed-off-by: David Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Acked-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SELinux: add security hook call to mediate attach_task (kernel/cpuset.c)
David Quigley [Fri, 23 Jun 2006 09:04:00 +0000 (02:04 -0700)]
[PATCH] SELinux: add security hook call to mediate attach_task (kernel/cpuset.c)

Add a security hook call to enable security modules to control the ability
to attach a task to a cpuset.  While limited control over this operation is
possible via permission checks on the pseudo fs interface, those checks are
not sufficient to control access to the target task, which is looked up in
this function.  The existing task_setscheduler hook is re-used for this
operation since this falls under the same class of operations.

Signed-off-by: David Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SELinux: add security hooks to {get,set}affinity
David Quigley [Fri, 23 Jun 2006 09:03:59 +0000 (02:03 -0700)]
[PATCH] SELinux: add security hooks to {get,set}affinity

This patch adds LSM hooks into the setaffinity and getaffinity functions to
enable security modules to control these operations between tasks with
task_setscheduler and task_getscheduler LSM hooks.

Signed-off-by: David Quigley <dpquigl@tycho.nsa.gov>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] lsm: add task_setioprio hook
James Morris [Fri, 23 Jun 2006 09:03:58 +0000 (02:03 -0700)]
[PATCH] lsm: add task_setioprio hook

Implement an LSM hook for setting a task's IO priority, similar to the hook
for setting a tasks's nice value.

A previous version of this LSM hook was included in an older version of
multiadm by Jan Engelhardt, although I don't recall it being submitted
upstream.

Also included is the corresponding SELinux hook, which re-uses the setsched
permission in the proccess class.

Signed-off-by: James Morris <jmorris@namei.org>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Cc: Jan Engelhardt <jengelh@linux01.gwdg.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Jens Axboe <axboe@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] move_pages: fix 32 -> 64 bit compat function
Christoph Lameter [Fri, 23 Jun 2006 09:03:57 +0000 (02:03 -0700)]
[PATCH] move_pages: fix 32 -> 64 bit compat function

The definition of the third parameter is a pointer to an array of virtual
addresses which give us some trouble.  The existing code calculated the
wrong address in the array since I used void to avoid having to specify a
type.

I now use the correct type "compat_uptr_t __user *" in the definition of
the function in kernel/compat.c.

However, I used __u32 in syscalls.h.  Would have to include compat.h there
in order to provide the same definition which would generate an ugly
include situation.

On both ia64 and x86_64 compat_uptr_t is u32. So this works although
parameter declarations differ.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sys_move_pages: 32bit support (i386, x86_64)
Christoph Lameter [Fri, 23 Jun 2006 09:03:56 +0000 (02:03 -0700)]
[PATCH] sys_move_pages: 32bit support (i386, x86_64)

sys_move_pages() support for 32bit (i386 plus x86_64 compat layer)

Add support for move_pages() on i386 and also add the compat functions
necessary to run 32 bit binaries on x86_64.

Add compat_sys_move_pages to the x86_64 32bit binary layer.  Note that it is
not up to date so I added the missing pieces.  Not sure if this is done the
right way.

[akpm@osdl.org: compile fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sys_move_pages: x86_64 support
Christoph Lameter [Fri, 23 Jun 2006 09:03:56 +0000 (02:03 -0700)]
[PATCH] sys_move_pages: x86_64 support

sys_move_pages support for x86_64

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: sys_move_pages(): support moving of individual pages
Christoph Lameter [Fri, 23 Jun 2006 09:03:55 +0000 (02:03 -0700)]
[PATCH] page migration: sys_move_pages(): support moving of individual pages

move_pages() is used to move individual pages of a process. The function can
be used to determine the location of pages and to move them onto the desired
node. move_pages() returns status information for each page.

long move_pages(pid, number_of_pages_to_move,
addresses_of_pages[],
nodes[] or NULL,
status[],
flags);

The addresses of pages is an array of void * pointing to the
pages to be moved.

The nodes array contains the node numbers that the pages should be moved
to. If a NULL is passed instead of an array then no pages are moved but
the status array is updated. The status request may be used to determine
the page state before issuing another move_pages() to move pages.

The status array will contain the state of all individual page migration
attempts when the function terminates. The status array is only valid if
move_pages() completed successfullly.

Possible page states in status[]:

0..MAX_NUMNODES The page is now on the indicated node.

-ENOENT Page is not present

-EACCES Page is mapped by multiple processes and can only
be moved if MPOL_MF_MOVE_ALL is specified.

-EPERM The page has been mlocked by a process/driver and
cannot be moved.

-EBUSY Page is busy and cannot be moved. Try again later.

-EFAULT Invalid address (no VMA or zero page).

-ENOMEM Unable to allocate memory on target node.

-EIO Unable to write back page. The page must be written
back in order to move it since the page is dirty and the
filesystem does not provide a migration function that
would allow the moving of dirty pages.

-EINVAL A dirty page cannot be moved. The filesystem does not provide
a migration function and has no ability to write back pages.

The flags parameter indicates what types of pages to move:

MPOL_MF_MOVE Move pages that are only mapped by the process.

MPOL_MF_MOVE_ALL Also move pages that are mapped by multiple processes.
Requires sufficient capabilities.

Possible return codes from move_pages()

-ENOENT No pages found that would require moving. All pages
are either already on the target node, not present, had an
invalid address or could not be moved because they were
mapped by multiple processes.

-EINVAL Flags other than MPOL_MF_MOVE(_ALL) specified or an attempt
to migrate pages in a kernel thread.

-EPERM MPOL_MF_MOVE_ALL specified without sufficient priviledges.
or an attempt to move a process belonging to another user.

-EACCES One of the target nodes is not allowed by the current cpuset.

-ENODEV One of the target nodes is not online.

-ESRCH Process does not exist.

-E2BIG Too many pages to move.

-ENOMEM Not enough memory to allocate control array.

-EFAULT Parameters could not be accessed.

A test program for move_pages() may be found with the patches
on ftp.kernel.org:/pub/linux/kernel/people/christoph/pmig/patches-2.6.17-rc4-mm3

From: Christoph Lameter <clameter@sgi.com>

  Detailed results for sys_move_pages()

  Pass a pointer to an integer to get_new_page() that may be used to
  indicate where the completion status of a migration operation should be
  placed.  This allows sys_move_pags() to report back exactly what happened to
  each page.

  Wish there would be a better way to do this. Looks a bit hacky.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Andi Kleen <ak@muc.de>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: use allocator function for migrate_pages()
Christoph Lameter [Fri, 23 Jun 2006 09:03:53 +0000 (02:03 -0700)]
[PATCH] page migration: use allocator function for migrate_pages()

Instead of passing a list of new pages, pass a function to allocate a new
page.  This allows the correct placement of MPOL_INTERLEAVE pages during page
migration.  It also further simplifies the callers of migrate pages.
migrate_pages() becomes similar to migrate_pages_to() so drop
migrate_pages_to().  The batching of new page allocations becomes unnecessary.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: handle freeing of pages in migrate_pages()
Christoph Lameter [Fri, 23 Jun 2006 09:03:52 +0000 (02:03 -0700)]
[PATCH] page migration: handle freeing of pages in migrate_pages()

Do not leave pages on the lists passed to migrate_pages().  Seems that we will
not need any postprocessing of pages.  This will simplify the handling of
pages by the callers of migrate_pages().

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: simplify migrate_pages()
Christoph Lameter [Fri, 23 Jun 2006 09:03:51 +0000 (02:03 -0700)]
[PATCH] page migration: simplify migrate_pages()

Currently migrate_pages() is mess with lots of goto.  Extract two functions
from migrate_pages() and get rid of the gotos.

Plus we can just unconditionally set the locked bit on the new page since we
are the only one holding a reference.  Locking is to stop others from
accessing the page once we establish references to the new page.

Remove the list_del from move_to_lru in order to have finer control over list
processing.

[akpm@osdl.org: add debug check]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] printk() should not be called under zone->lock
Kirill Korotaev [Fri, 23 Jun 2006 09:03:50 +0000 (02:03 -0700)]
[PATCH] printk() should not be called under zone->lock

This patch fixes printk() under zone->lock in show_free_areas().  It can be
unsafe to call printk() under this lock, since caller can try to
allocate/free some memory and selfdeadlock on this lock.  I found
allocations/freeing mem both in netconsole and serial console.

This issue was faced in reallity when meminfo was periodically printed for
debug purposes and netconsole was used.

Signed-off-by: Kirill Korotaev <dev@openvz.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Delete unused definitions of kvaddr_to_nid
Ralf Baechle [Fri, 23 Jun 2006 09:03:50 +0000 (02:03 -0700)]
[PATCH] Delete unused definitions of kvaddr_to_nid

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kernel-doc for mm/filemap.c
Randy Dunlap [Fri, 23 Jun 2006 09:03:49 +0000 (02:03 -0700)]
[PATCH] kernel-doc for mm/filemap.c

mm/filemap.c:
- add lots of kernel-doc;
- fix some typos and kernel-doc errors;
- drop some blank lines between function close and EXPORT_SYMBOL();

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: kmalloc, kzalloc comments cleanup and fix
Paul Drynoff [Fri, 23 Jun 2006 09:03:48 +0000 (02:03 -0700)]
[PATCH] slab: kmalloc, kzalloc comments cleanup and fix

- Move comments for kmalloc to right place, currently it near __do_kmalloc

- Comments for kzalloc

- More detailed comments for kmalloc

- Appearance of "kmalloc" and "kzalloc" man pages after "make mandocs"

[rdunlap@xenotime.net: simplification]
Signed-off-by: Paul Drynoff <pauldrynoff@gmail.com>
Acked-by: Randy Dunlap <rdunlap@xenotime.net>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] update vm_total_pages at memory hotadd
KAMEZAWA Hiroyuki [Fri, 23 Jun 2006 09:03:47 +0000 (02:03 -0700)]
[PATCH] update vm_total_pages at memory hotadd

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] initialise total_memory() earlier
Andrew Morton [Fri, 23 Jun 2006 09:03:47 +0000 (02:03 -0700)]
[PATCH] initialise total_memory() earlier

Initialise total_memory earlier in boot.  Because if for some reason we run
page reclaim early in boot, we don't want total_memory to be zero when we use
it as a divisor.

And rename total_memory to vm_total_pages to avoid naming clashes with
architectures.

Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Martin Bligh <mbligh@google.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm/slab.c: fix early init assumption
Ingo Molnar [Fri, 23 Jun 2006 09:03:46 +0000 (02:03 -0700)]
[PATCH] mm/slab.c: fix early init assumption

The SLAB bootstrap code assumes that the first two kmalloc caches created
(the INDEX_AC and INDEX_L3 kmalloc caches) wont be off-slab.  But due to AC
and L3 structure size increase in lockdep, one of them ended up being
off-slab, and subsequently crashing with:

Unable to handle kernel NULL pointer dereference at 0000000000000000 RIP:
 [<ffffffff80267478>] kmem_cache_alloc+0x26/0x7d

The fix is to introduce a bootstrap flag and to use it to prevent off-slab
caches being created so early during bootup.

(The calculation for off-slab caches is quite complex so i didnt want to
complicate things with introducing yet another INDEX_ calculation, the flag
approach is simpler and smaller.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix update_mmu_cache in fremap.c
Hugh Dickins [Fri, 23 Jun 2006 09:03:45 +0000 (02:03 -0700)]
[PATCH] fix update_mmu_cache in fremap.c

There are two calls to update_mmu_cache in fremap.c, both defective.
The one in install_page needs to be accompanied by lazy_mmu_prot_update
(some other cleanup time, move that into ia64 update_mmu_cache itself); and
the one in install_file_pte should be removed since the pte is not present.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] remove unused o_flags from do_shmat
Hugh Dickins [Fri, 23 Jun 2006 09:03:45 +0000 (02:03 -0700)]
[PATCH] remove unused o_flags from do_shmat

Remove the unused variable o_flags from do_shmat.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swapoff: use atomic_inc_not_zero() on mm_users
Hugh Dickins [Fri, 23 Jun 2006 09:03:44 +0000 (02:03 -0700)]
[PATCH] swapoff: use atomic_inc_not_zero() on mm_users

Now that we have atomic_inc_not_zero, it's more elegant for try_to_unuse to
use that on mm_users: doesn't actually matter at present, but safer to be
sure that once mm_users has gone to 0, nothing raises it for an instant.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] add page_mkwrite() vm_operations method
David Howells [Fri, 23 Jun 2006 09:03:43 +0000 (02:03 -0700)]
[PATCH] add page_mkwrite() vm_operations method

Add a new VMA operation to notify a filesystem or other driver about the
MMU generating a fault because userspace attempted to write to a page
mapped through a read-only PTE.

This facility permits the filesystem or driver to:

 (*) Implement storage allocation/reservation on attempted write, and so to
     deal with problems such as ENOSPC more gracefully (perhaps by generating
     SIGBUS).

 (*) Delay making the page writable until the contents have been written to a
     backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS.
     It permits the filesystem to have some guarantee about the state of the
     cache.

 (*) Account and limit number of dirty pages. This is one piece of the puzzle
     needed to make shared writable mapping work safely in FUSE.

Needed by cachefs (Or is it cachefiles?  Or fscache? <head spins>).

At least four other groups have stated an interest in it or a desire to use
the functionality it provides: FUSE, OCFS2, NTFS and JFFS2.  Also, things like
EXT3 really ought to use it to deal with the case of shared-writable mmap
encountering ENOSPC before we permit the page to be dirtied.

From: Peter Zijlstra <a.p.zijlstra@chello.nl>

  get_user_pages(.write=1, .force=1) can generate COW hits on read-only
  shared mappings, this patch traps those as mkpage_write candidates and fails
  to handle them the old way.

Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Joel Becker <Joel.Becker@oracle.com>
Cc: Mark Fasheh <mark.fasheh@oracle.com>
Cc: Anton Altaparmakov <aia21@cantab.net>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: fix swap unused warning
Con Kolivas [Fri, 23 Jun 2006 09:03:42 +0000 (02:03 -0700)]
[PATCH] mm: fix swap unused warning

If CONFIG_SWAP is not defined we get:

mm/vmscan.c: In function Ã¢\80\98remove_mappingâ\80\99:
mm/vmscan.c:387: warning: unused variable Ã¢\80\98swapâ\80\99

Convert defines in swap.h into blank inline functions to fix this warning
and be consistent.

Signed-off-by: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sparsemem: record nid during memory present
Andy Whitcroft [Fri, 23 Jun 2006 09:03:41 +0000 (02:03 -0700)]
[PATCH] sparsemem: record nid during memory present

Record the node id as we mark sections for instantiation.  Use this nid
during instantiation to direct allocations.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Mike Kravetz <kravetz@us.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Bob Picco <bob.picco@hp.com>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Martin Bligh <mbligh@google.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: verify pointers before free
Pekka Enberg [Fri, 23 Jun 2006 09:03:40 +0000 (02:03 -0700)]
[PATCH] slab: verify pointers before free

Passing an invalid pointer to kfree() and kmem_cache_free() is likely to
cause bad memory corruption or even take down the whole system because the
bad pointer is likely reused immediately due to the per-CPU caches.  Until
now, we don't do any verification for this if CONFIG_DEBUG_SLAB is
disabled.

As suggested by Linus, add PageSlab check to page_to_cache() and
page_to_slab() to verify pointers passed to kfree().  Also, move the
stronger check from cache_free_debugcheck() to kmem_cache_free() to ensure
the passed pointer actually belongs to the cache we're about to free the
object.

For page_to_cache() and page_to_slab(), the assertions should have
virtually no extra cost (two instructions, no data cache pressure) and for
kmem_cache_free() the overhead should be minimal.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: Update documentation
Christoph Lameter [Fri, 23 Jun 2006 09:03:39 +0000 (02:03 -0700)]
[PATCH] page migration: Update documentation

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] More page migration: use migration entries for file pages
Christoph Lameter [Fri, 23 Jun 2006 09:03:38 +0000 (02:03 -0700)]
[PATCH] More page migration: use migration entries for file pages

This implements the use of migration entries to preserve ptes of file backed
pages during migration.  Processes can therefore be migrated back and forth
without loosing their connection to pagecache pages.

Note that we implement the migration entries only for linear mappings.
Nonlinear mappings still require the unmapping of the ptes for migration.

And another writepage() ugliness shows up.  writepage() can drop the page
lock.  Therefore we have to remove migration ptes before calling writepages()
in order to avoid having migration entries point to unlocked pages.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] More page migration: do not inc/dec rss counters
Christoph Lameter [Fri, 23 Jun 2006 09:03:38 +0000 (02:03 -0700)]
[PATCH] More page migration: do not inc/dec rss counters

If we install a migration entry then the rss not really decreases since the
page is just moved somewhere else.  We can save ourselves the work of
decrementing and later incrementing which will just eventually cause cacheline
bouncing.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Swapless page migration: modify core logic
Christoph Lameter [Fri, 23 Jun 2006 09:03:37 +0000 (02:03 -0700)]
[PATCH] Swapless page migration: modify core logic

Use the migration entries for page migration

This modifies the migration code to use the new migration entries.  It now
becomes possible to migrate anonymous pages without having to add a swap
entry.

We add a couple of new functions to replace migration entries with the proper
ptes.

We cannot take the tree_lock for migrating anonymous pages anymore.  However,
we know that we hold the only remaining reference to the page when the page
count reaches 1.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Swapless page migration: rip out swap based logic
Christoph Lameter [Fri, 23 Jun 2006 09:03:36 +0000 (02:03 -0700)]
[PATCH] Swapless page migration: rip out swap based logic

Rip the page migration logic out.

Remove all code that has to do with swapping during page migration.

This also guts the ability to migrate pages to swap.  No one used that so lets
let it go for good.

Page migration should be a bit broken after this patch.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Swapless page migration: add R/W migration entries
Christoph Lameter [Fri, 23 Jun 2006 09:03:35 +0000 (02:03 -0700)]
[PATCH] Swapless page migration: add R/W migration entries

Implement read/write migration ptes

We take the upper two swapfiles for the two types of migration ptes and define
a series of macros in swapops.h.

The VM is modified to handle the migration entries.  migration entries can
only be encountered when the page they are pointing to is locked.  This limits
the number of places one has to fix.  We also check in copy_pte_range and in
mprotect_pte_range() for migration ptes.

We check for migration ptes in do_swap_cache and call a function that will
then wait on the page lock.  This allows us to effectively stop all accesses
to apge.

Migration entries are created by try_to_unmap if called for migration and
removed by local functions in migrate.c

From: Hugh Dickins <hugh@veritas.com>

  Several times while testing swapless page migration (I've no NUMA, just
  hacking it up to migrate recklessly while running load), I've hit the
  BUG_ON(!PageLocked(p)) in migration_entry_to_page.

  This comes from an orphaned migration entry, unrelated to the current
  correctly locked migration, but hit by remove_anon_migration_ptes as it
  checks an address in each vma of the anon_vma list.

  Such an orphan may be left behind if an earlier migration raced with fork:
  copy_one_pte can duplicate a migration entry from parent to child, after
  remove_anon_migration_ptes has checked the child vma, but before it has
  removed it from the parent vma.  (If the process were later to fault on this
  orphaned entry, it would hit the same BUG from migration_entry_wait.)

  This could be fixed by locking anon_vma in copy_one_pte, but we'd rather
  not.  There's no such problem with file pages, because vma_prio_tree_add
  adds child vma after parent vma, and the page table locking at each end is
  enough to serialize.  Follow that example with anon_vma: add new vmas to the
  tail instead of the head.

  (There's no corresponding problem when inserting migration entries,
  because a missed pte will leave the page count and mapcount high, which is
  allowed for.  And there's no corresponding problem when migrating via swap,
  because a leftover swap entry will be correctly faulted.  But the swapless
  method has no refcounting of its entries.)

From: Ingo Molnar <mingo@elte.hu>

  pte_unmap_unlock() takes the pte pointer as an argument.

From: Hugh Dickins <hugh@veritas.com>

  Several times while testing swapless page migration, gcc has tried to exec
  a pointer instead of a string: smells like COW mappings are not being
  properly write-protected on fork.

  The protection in copy_one_pte looks very convincing, until at last you
  realize that the second arg to make_migration_entry is a boolean "write",
  and SWP_MIGRATION_READ is 30.

  Anyway, it's better done like in change_pte_range, using
  is_write_migration_entry and make_migration_entry_read.

From: Hugh Dickins <hugh@veritas.com>

  Remove unnecessary obfuscation from sys_swapon's range check on swap type,
  which blew up causing memory corruption once swapless migration made
  MAX_SWAPFILES no longer 2 ^ MAX_SWAPFILES_SHIFT.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
From: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: move fallback handling into special function
Christoph Lameter [Fri, 23 Jun 2006 09:03:33 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: move fallback handling into special function

Move the fallback code into a new fallback function and make the function
behave like any other migration function.  This requires retaking the lock if
pageout() drops it.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: pass "mapping" to migration functions
Christoph Lameter [Fri, 23 Jun 2006 09:03:33 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: pass "mapping" to migration functions

Change handling of address spaces.

Pass a pointer to the address space in which the page is migrated to all
migration function.  This avoids repeatedly having to retrieve the address
space pointer from the page and checking it for validity.  The old page
mapping will change once migration has gone to a certain step, so it is less
confusing to have the pointer always available.

Move the setting of the mapping and index for the new page into
migrate_pages().

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: extract try_to_unmap from migration functions
Christoph Lameter [Fri, 23 Jun 2006 09:03:32 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: extract try_to_unmap from migration functions

Extract try_to_unmap and rename remove_references -> move_mapping

try_to_unmap() may significantly change the page state by for example setting
the dirty bit.  It is therefore best to unmap in migrate_pages() before
calling any migration functions.

migrate_page_remove_references() will then only move the new page in place of
the old page in the mapping.  Rename the function to
migrate_page_move_mapping().

This allows us to get rid of the special unmapping for the fallback path.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: drop nr_refs in remove_references()
Christoph Lameter [Fri, 23 Jun 2006 09:03:29 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: drop nr_refs in remove_references()

Drop nr_refs parameter from migrate_page_remove_references()

The nr_refs parameter is not really useful since the number of remaining
references is always

1 for anonymous pages without a mapping
2 for pages with a mapping
3 for pages with a mapping and PagePrivate set.

Remove the early check for the number of references since we are checking
page_mapcount() earlier.  Ultimately only the refcount matters after the
tree_lock has been obtained.

Signed-off-by: Christoph Lameter <clameter@sgi.coim>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: remove useless definitions
Christoph Lameter [Fri, 23 Jun 2006 09:03:29 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: remove useless definitions

Remove the export for migrate_page_remove_references() and migrate_page_copy()
that are unlikely to be used directly by filesystems implementing migration.
The export was useful when buffer_migrate_page() lived in fs/buffer.c but it
has now been moved to migrate.c in the migration reorg.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: group functions
Christoph Lameter [Fri, 23 Jun 2006 09:03:28 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: group functions

Reorder functions in migrate.c.  Group all migration functions for struct
address_space_operations together.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration cleanup: rename "ignrefs" to "migration"
Christoph Lameter [Fri, 23 Jun 2006 09:03:27 +0000 (02:03 -0700)]
[PATCH] page migration cleanup: rename "ignrefs" to "migration"

migrate is a better name since it is only used by page migration.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] writeback: fix range handling
OGAWA Hirofumi [Fri, 23 Jun 2006 09:03:26 +0000 (02:03 -0700)]
[PATCH] writeback: fix range handling

When a writeback_control's `start' and `end' fields are used to
indicate a one-byte-range starting at file offset zero, the required
values of .start=0,.end=0 mean that the ->writepages() implementation
has no way of telling that it is being asked to perform a range
request.  Because we're currently overloading (start == 0 && end == 0)
to mean "this is not a write-a-range request".

To make all this sane, the patch changes range of writeback_control.

So caller does: If it is calling ->writepages() to write pages, it
sets range (range_start/end or range_cyclic) always.

And if range_cyclic is true, ->writepages() thinks the range is
cyclic, otherwise it just uses range_start and range_end.

This patch does,

    - Add LLONG_MAX, LLONG_MIN, ULLONG_MAX to include/linux/kernel.h
      -1 is usually ok for range_end (type is long long). But, if someone did,

range_end += val; range_end is "val - 1"
u64val = range_end >> bits; u64val is "~(0ULL)"

      or something, they are wrong. So, this adds LLONG_MAX to avoid nasty
      things, and uses LLONG_MAX for range_end.

    - All callers of ->writepages() sets range_start/end or range_cyclic.

    - Fix updates of ->writeback_index. It seems already bit strange.
      If it starts at 0 and ended by check of nr_to_write, this last
      index may reduce chance to scan end of file.  So, this updates
      ->writeback_index only if range_cyclic is true or whole-file is
      scanned.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Anton Altaparmakov <aia21@cantab.net>
Cc: Steven French <sfrench@us.ibm.com>
Cc: "Vladimir V. Saveliev" <vs@namesys.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] buglet in radix_tree_tag_set
Peter Zijlstra [Fri, 23 Jun 2006 09:03:25 +0000 (02:03 -0700)]
[PATCH] buglet in radix_tree_tag_set

The comment states: 'Setting a tag on a not-present item is a BUG.' Hence
if 'index' is larger than the maxindex; the item _cannot_ be presen; it
should also be a BUG.

Also, this allows the following statement (assume a fresh tree):

  radix_tree_tag_set(root, 16, 1);

to fail silently, but when preceded by:

  radix_tree_insert(root, 32, item);

it would BUG, because the height has been extended by the insert.

In neither case was 16 present.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: redzone double-free detection
Pekka Enberg [Fri, 23 Jun 2006 09:03:24 +0000 (02:03 -0700)]
[PATCH] slab: redzone double-free detection

At present our slab debugging tells us that it detected a double-free or
corruption - it does not distinguish between them.  Sometimes it's useful
to be able to differentiate between these two types of information.

Add double-free detection to redzone verification when freeing an object.
As explained by Manfred, when we are freeing an object, both redzones
should be RED_ACTIVE.  However, if both are RED_INACTIVE, we are trying to
free an object that was already free'd.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] likely cleanup: remove unlikely in sys_mprotect()
Hua Zhong [Fri, 23 Jun 2006 09:03:23 +0000 (02:03 -0700)]
[PATCH] likely cleanup: remove unlikely in sys_mprotect()

With likely/unlikely profiling on my not-so-busy-typical-developmentsystem
there are 5k misses vs 2k hits.  So I guess we should remove the unlikely.

Signed-off-by: Hua Zhong <hzhong@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] radix-tree: small
Nick Piggin [Fri, 23 Jun 2006 09:03:22 +0000 (02:03 -0700)]
[PATCH] radix-tree: small

Reduce radix tree node memory usage by about a factor of 4 for small files
(< 64K).  There are pointer traversal and memory usage costs for large
files with dense pagecache.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] radix-tree: direct data
Nick Piggin [Fri, 23 Jun 2006 09:03:22 +0000 (02:03 -0700)]
[PATCH] radix-tree: direct data

The ability to have height 0 radix trees (a direct pointer to the data item
rather than going through a full node->slot) quietly disappeared with
old-2.6-bkcvs commit ffee171812d51652f9ba284302d9e5c5cc14bdfd.  On 64-bit
machines this causes nearly 600 bytes to be used for every <= 4K file in
pagecache.

Re-introduce this feature, root tags stored in spare ->gfp_mask bits.

Simplify radix_tree_delete's complex tag clearing arrangement (which would
become even more complex) by just falling back to tag clearing functions
(the pagecache radix-tree never uses this path anyway, so the icache
savings will mean it's actually a speedup).

On my 4GB G5, this saves 8MB RAM per kernel kernel source+object tree in
pagecache.

Pagecache lookup, insertion, and removal speed for small files will also be
improved.

This makes RCU radix tree harder, but it's worth it.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] change gen_pool allocator to not touch managed memory
Dean Nelson [Fri, 23 Jun 2006 09:03:21 +0000 (02:03 -0700)]
[PATCH] change gen_pool allocator to not touch managed memory

Modify the gen_pool allocator (lib/genalloc.c) to utilize a bitmap scheme
instead of the buddy scheme.  The purpose of this change is to eliminate
the touching of the actual memory being allocated.

Since the change modifies the interface, a change to the uncached allocator
(arch/ia64/kernel/uncached.c) is also required.

Both Andrey Volkov and Jes Sorenson have expressed a desire that the
gen_pool allocator not write to the memory being managed. See the
following:

  http://marc.theaimsgroup.com/?l=linux-kernel&m=113518602713125&w=2
  http://marc.theaimsgroup.com/?l=linux-kernel&m=113533568827916&w=2

Signed-off-by: Dean Nelson <dcn@sgi.com>
Cc: Andrey Volkov <avolkov@varma-el.com>
Acked-by: Jes Sorensen <jes@trained-monkey.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: introduce remap_vmalloc_range()
Nick Piggin [Fri, 23 Jun 2006 09:03:20 +0000 (02:03 -0700)]
[PATCH] mm: introduce remap_vmalloc_range()

Add remap_vmalloc_range, vmalloc_user, and vmalloc_32_user so that drivers
can have a nice interface for remapping vmalloc memory.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Unify pxm_to_node() and node_to_pxm()
Yasunori Goto [Fri, 23 Jun 2006 09:03:19 +0000 (02:03 -0700)]
[PATCH] Unify pxm_to_node() and node_to_pxm()

Consolidate the various arch-specific implementations of pxm_to_node() and
node_to_pxm() into a single generic version.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@muc.de>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: "Brown, Len" <len.brown@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: rework memory shrinker
Rafael J. Wysocki [Fri, 23 Jun 2006 09:03:18 +0000 (02:03 -0700)]
[PATCH] swsusp: rework memory shrinker

Rework the swsusp's memory shrinker in the following way:

- Simplify balance_pgdat() by removing all of the swsusp-related code
  from it.

- Make shrink_all_memory() use shrink_slab() and a new function
  shrink_all_zones() which calls shrink_active_list() and
  shrink_inactive_list() directly for each zone in a way that's optimized
  for suspend.

In shrink_all_memory() we try to free exactly as many pages as the caller
asks for, preferably in one shot, starting from easier targets.  Â If slab
caches are huge, they are most likely to have enough pages to reclaim.
 The inactive lists are next (the zones with more inactive pages go first)
etc.

Each time shrink_all_memory() attempts to shrink the active and inactive
lists for each zone in 5 passes.  Â In the first pass, only the inactive
lists are taken into consideration.  Â In the next two passes the active
lists are also shrunk, but mapped pages are not reclaimed.  Â In the last
two passes the active and inactive lists are shrunk and mapped pages are
reclaimed as well.  The aim of this is to alter the reclaim logic to choose
the best pages to keep on resume and improve the responsiveness of the
resumed system.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: stop using list_for_each
Christoph Hellwig [Fri, 23 Jun 2006 09:03:17 +0000 (02:03 -0700)]
[PATCH] slab: stop using list_for_each

Use the _entry variant everywhere to clean the code up a tiny bit.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: clean up kmem_getpages
Christoph Hellwig [Fri, 23 Jun 2006 09:03:17 +0000 (02:03 -0700)]
[PATCH] slab: clean up kmem_getpages

The last ifdef addition hit the ugliness treshold on this functions, so:

 - rename the variable i to nr_pages so it's somewhat descriptive
 - remove the addr variable and do the page_address call at the very end
 - instead of ifdef'ing the whole alloc_pages_node call just make the
   __GFP_COMP addition to flags conditional
 - rewrite the __GFP_COMP comment to make sense

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] tightening hugetlb strict accounting
Chen, Kenneth W [Fri, 23 Jun 2006 09:03:15 +0000 (02:03 -0700)]
[PATCH] tightening hugetlb strict accounting

Current hugetlb strict accounting for shared mapping always assume mapping
starts at zero file offset and reserves pages between zero and size of the
file.  This assumption often reserves (or lock down) a lot more pages then
necessary if application maps at none zero file offset.  libhugetlbfs is
one example that requires proper reservation on shared mapping starts at
none zero offset.

This patch extends the reservation and hugetlb strict accounting to support
any arbitrary pair of (offset, len), resulting a much more robust and
accurate scheme.  More importantly, it won't lock down any hugetlb pages
outside file mapping.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Acked-by: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] reserve space for swap label
Andreas Dilger [Fri, 23 Jun 2006 09:03:14 +0000 (02:03 -0700)]
[PATCH] reserve space for swap label

Reserve space in the swap disk header for a LABEL and UUID to be specified.
 This has been possible with util-linux-2.12b (via e2fsprogs 1.36
libblkid), and is used by at least FC3 and later.  The kernel doesn't
really care about this, but the space shouldn't accidentally be used by
something else either.

Also make the on-disk structures be fixed-size types, instead of "int",
though I don't know of any architecture in use where an "int" isn't the
same size as a "__u32" (all current kernel arches have it as "unsigned
int").

Signed-off-by: Andreas Dilger <adilger@shaw.ca>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: fix typos in comments in mm/oom_kill.c
Dave Peterson [Fri, 23 Jun 2006 09:03:13 +0000 (02:03 -0700)]
[PATCH] mm: fix typos in comments in mm/oom_kill.c

This fixes a few typos in the comments in mm/oom_kill.c.

Signed-off-by: David S. Peterson <dsp@llnl.gov>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] support for panic at OOM
KAMEZAWA Hiroyuki [Fri, 23 Jun 2006 09:03:13 +0000 (02:03 -0700)]
[PATCH] support for panic at OOM

This patch adds panic_on_oom sysctl under sys.vm.

When sysctl vm.panic_on_oom = 1, the kernel panics intead of killing rogue
processes.  And if vm.panic_on_oom is 0 the kernel will do oom_kill() in
the same way as it does today.  Of course, the default value is 0 and only
root can modifies it.

In general, oom_killer works well and kill rogue processes.  So the whole
system can survive.  But there are environments where panic is preferable
rather than kill some processes.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] squash duplicate page_to_pfn and pfn_to_page
Andy Whitcroft [Fri, 23 Jun 2006 09:03:12 +0000 (02:03 -0700)]
[PATCH] squash duplicate page_to_pfn and pfn_to_page

We have architectures where the size of page_to_pfn and pfn_to_page are
significant enough to overall image size that they wish to push them out of
line.  However, in the process we have grown a second copy of the
implementation of each of these routines for each memory model.  Share the
implmentation exposing it either inline or out-of-line as required.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] wait_table and zonelist initializing for memory hotadd: update zonelists
Yasunori Goto [Fri, 23 Jun 2006 09:03:11 +0000 (02:03 -0700)]
[PATCH] wait_table and zonelist initializing for memory hotadd: update zonelists

In current code, zonelist is considered to be build once, no modification.
But MemoryHotplug can add new zone/pgdat.  It must be updated.

This patch modifies build_all_zonelists().  By this, build_all_zonelist() can
reconfig pgdat's zonelists.

To update them safety, this patch use stop_machine_run().  Other cpus don't
touch among updating them by using it.

In old version (V2 of node hotadd), kernel updated them after zone
initialization.  But present_page of its new zone is still 0, because
online_page() is not called yet at this time.  Build_zonelists() checks
present_pages to find present zone.  It was too early.  So, I changed it after
online_pages().

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] wait_table and zonelist initializing for memory hotadd: wait_table initialization
Yasunori Goto [Fri, 23 Jun 2006 09:03:10 +0000 (02:03 -0700)]
[PATCH] wait_table and zonelist initializing for memory hotadd: wait_table initialization

Wait_table is initialized according to zone size at boot time.  But, we cannot
know the maixmum zone size when memory hotplug is enabled.  It can be
changed....  And resizing of wait_table is hard.

So kernel allocate and initialzie wait_table as its maximum size.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] wait_table and zonelist initializing for memory hotadd: add return code for...
Yasunori Goto [Fri, 23 Jun 2006 09:03:10 +0000 (02:03 -0700)]
[PATCH] wait_table and zonelist initializing for memory hotadd: add return code for init_current_empty_zone

When add_zone() is called against empty zone (not populated zone), we have to
initialize the zone which didn't initialize at boot time.  But,
init_currently_empty_zone() may fail due to allocation of wait table.  So,
this patch is to catch its error code.

Changes against wait_table is in the next patch.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] wait_table and zonelist initializing for memory hotadd: change to meminit...
Yasunori Goto [Fri, 23 Jun 2006 09:03:09 +0000 (02:03 -0700)]
[PATCH] wait_table and zonelist initializing for memory hotadd: change to meminit for build_zonelist

Change definitions of some functions and data from __init to __meminit.

These functions and data can be used after bootup by this patch to be used for
hot-add codes.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] wait_table and zonelist initializing for memory hotadd: change name of wait_t...
Yasunori Goto [Fri, 23 Jun 2006 09:03:08 +0000 (02:03 -0700)]
[PATCH] wait_table and zonelist initializing for memory hotadd: change name of wait_table_size()

This is just to rename from wait_table_size() to wait_table_hash_nr_entries().

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] migration: remove unnecessary PageSwapCache checks
Christoph Lameter [Fri, 23 Jun 2006 09:03:08 +0000 (02:03 -0700)]
[PATCH] migration: remove unnecessary PageSwapCache checks

Remove two unnecessary PageSwapCache checks.  The page refcount is raised
and therefore page migration cannot occur in both functions.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: page mapping cleanup
Pekka Enberg [Fri, 23 Jun 2006 09:03:07 +0000 (02:03 -0700)]
[PATCH] slab: page mapping cleanup

Clean up slab allocator page mapping a bit.  The memory allocated for a
slab is physically contiguous so it is okay to assume struct pages are too
so kill the long-standing comment.  Furthermore, rename set_slab_attr to
slab_map_pages and add a comment explaining why its needed.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PG_uncached is ia64 only
Andrew Morton [Fri, 23 Jun 2006 09:03:06 +0000 (02:03 -0700)]
[PATCH] PG_uncached is ia64 only

As Nick points out, only ia64 uses PG_uncached.  So we can push it up into the
higher bits of the lower half of page->flags and make room for another flag on
32-bit machines.

Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Jesse Barnes <jbarnes@sgi.com>
Cc: Jes Sorensen <jes@trained-monkey.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: extract cache_free_alien from __cache_free
Pekka Enberg [Fri, 23 Jun 2006 09:03:05 +0000 (02:03 -0700)]
[PATCH] slab: extract cache_free_alien from __cache_free

Move alien object freeing to cache_free_alien() to reduce #ifdef clutter in
__cache_free().

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Page Migration: Make do_swap_page redo the fault
Christoph Lameter [Fri, 23 Jun 2006 09:03:04 +0000 (02:03 -0700)]
[PATCH] Page Migration: Make do_swap_page redo the fault

It is better to redo the complete fault if do_swap_page() finds that the
page is not in PageSwapCache() because the page migration code may have
replaced the swap pte already with a pte pointing to valid memory.

do_swap_page() may interpret an invalid swap entry without this patch
because we do not reload the pte if we are looping back.  The page
migration code may already have reused the swap entry referenced by our
local swp_entry.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] zone handle unaligned zone boundaries
Andy Whitcroft [Fri, 23 Jun 2006 09:03:01 +0000 (02:03 -0700)]
[PATCH] zone handle unaligned zone boundaries

The buddy allocator has a requirement that boundaries between contigious
zones occur aligned with the the MAX_ORDER ranges.  Where they do not we
will incorrectly merge pages cross zone boundaries.  This can lead to pages
from the wrong zone being handed out.

Originally the buddy allocator would check that buddies were in the same
zone by referencing the zone start and end page frame numbers.  This was
removed as it became very expensive and the buddy allocator already made
the assumption that zones boundaries were aligned.

It is clear that not all configurations and architectures are honouring
this alignment requirement.  Therefore it seems safest to reintroduce
support for non-aligned zone boundaries.  This patch introduces a new check
when considering a page a buddy it compares the zone_table index for the
two pages and refuses to merge the pages where they do not match.  The
zone_table index is unique for each node/zone combination when
FLATMEM/DISCONTIGMEM is enabled and for each section/zone combination when
SPARSEMEM is enabled (a SPARSEMEM section is at least a MAX_ORDER size).

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] for_each_possible_cpu: xfs
KAMEZAWA Hiroyuki [Fri, 23 Jun 2006 09:03:00 +0000 (02:03 -0700)]
[PATCH] for_each_possible_cpu: xfs

for_each_cpu() actually iterates across all possible CPUs.  We've had mistakes
in the past where people were using for_each_cpu() where they should have been
iterating across only online or present CPUs.  This is inefficient and
possibly buggy.

We're renaming for_each_cpu() to for_each_possible_cpu() to avoid this in the
future.

This patch replaces for_each_cpu with for_each_possible_cpu.
in xfs.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Nathan Scott <nathans@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] XFS: Use the dentry passed to statfs() to limit the scope of the results
David Howells [Fri, 23 Jun 2006 09:02:59 +0000 (02:02 -0700)]
[PATCH] XFS: Use the dentry passed to statfs() to limit the scope of the results

Enable XFS to limit the statfs() results to the project quota covering the
dentry used as a base for call.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Nathan Scott <nathans@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] VFS: Permit filesystem to perform statfs with a known root dentry
David Howells [Fri, 23 Jun 2006 09:02:58 +0000 (02:02 -0700)]
[PATCH] VFS: Permit filesystem to perform statfs with a known root dentry

Give the statfs superblock operation a dentry pointer rather than a superblock
pointer.

This complements the get_sb() patch.  That reduced the significance of
sb->s_root, allowing NFS to place a fake root there.  However, NFS does
require a dentry to use as a target for the statfs operation.  This permits
the root in the vfsmount to be used instead.

linux/mount.h has been added where necessary to make allyesconfig build
successfully.

Interest has also been expressed for use with the FUSE and XFS filesystems.

Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] VFS: Permit filesystem to override root dentry on mount
David Howells [Fri, 23 Jun 2006 09:02:57 +0000 (02:02 -0700)]
[PATCH] VFS: Permit filesystem to override root dentry on mount

Extend the get_sb() filesystem operation to take an extra argument that
permits the VFS to pass in the target vfsmount that defines the mountpoint.

The filesystem is then required to manually set the superblock and root dentry
pointers.  For most filesystems, this should be done with simple_set_mnt()
which will set the superblock pointer and then set the root dentry to the
superblock's s_root (as per the old default behaviour).

The get_sb() op now returns an integer as there's now no need to return the
superblock pointer.

This patch permits a superblock to be implicitly shared amongst several mount
points, such as can be done with NFS to avoid potential inode aliasing.  In
such a case, simple_set_mnt() would not be called, and instead the mnt_root
and mnt_sb would be set directly.

The patch also makes the following changes:

 (*) the get_sb_*() convenience functions in the core kernel now take a vfsmount
     pointer argument and return an integer, so most filesystems have to change
     very little.

 (*) If one of the convenience function is not used, then get_sb() should
     normally call simple_set_mnt() to instantiate the vfsmount. This will
     always return 0, and so can be tail-called from get_sb().

 (*) generic_shutdown_super() now calls shrink_dcache_sb() to clean up the
     dcache upon superblock destruction rather than shrink_dcache_anon().

     This is required because the superblock may now have multiple trees that
     aren't actually bound to s_root, but that still need to be cleaned up. The
     currently called functions assume that the whole tree is rooted at s_root,
     and that anonymous dentries are not the roots of trees which results in
     dentries being left unculled.

     However, with the way NFS superblock sharing are currently set to be
     implemented, these assumptions are violated: the root of the filesystem is
     simply a dummy dentry and inode (the real inode for '/' may well be
     inaccessible), and all the vfsmounts are rooted on anonymous[*] dentries
     with child trees.

     [*] Anonymous until discovered from another tree.

 (*) The documentation has been adjusted, including the additional bit of
     changing ext2_* into foo_* in the documentation.

[akpm@osdl.org: convert ipath_fs, do other stuff]
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix cdrom being confused on using kdump
Rachita Kothiyal [Fri, 23 Jun 2006 09:02:56 +0000 (02:02 -0700)]
[PATCH] Fix cdrom being confused on using kdump

I have seen the cdrom drive appearing confused on using kdump on certain
x86_64 systems.  During the booting up of the second kernel, the following
message would keep flooding the console, and the booting would not proceed
any further.

hda: cdrom_pc_intr: The drive appears confused (ireason = 0x01)

In this patch, whenever we are hitting a confused state in the interrupt
handler with the DRQ set, we end the request and return ide_stopped.  Using
this I dont see the status error.

Signed-off-by: Rachita Kothiyal <rachita@in.ibm.com>
Acked-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/usb-2.6
Linus Torvalds [Fri, 23 Jun 2006 06:09:42 +0000 (23:09 -0700)]
Merge /pub/scm/linux/kernel/git/gregkh/usb-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/usb-2.6:
  [PATCH] Driver core: fix locking issues with the devices that are attached to classes
  [PATCH] USB: get USB suspend to work again

18 years ago[PATCH] Driver core: fix locking issues with the devices that are attached to classes
Greg Kroah-Hartman [Fri, 23 Jun 2006 00:17:32 +0000 (17:17 -0700)]
[PATCH] Driver core: fix locking issues with the devices that are attached to classes

Doh, that was foolish...

Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] USB: get USB suspend to work again
Greg Kroah-Hartman [Thu, 22 Jun 2006 20:29:52 +0000 (13:29 -0700)]
[PATCH] USB: get USB suspend to work again

Yeah, it's a hack, but it is only temporary until Alan's patches
reworking this area make it in.  We really should not care what devices
below us are doing, especially when we do not really know what type of
devices they are.  This patch relies on the fact that the endpoint
devices do not have a driver assigned to us.

Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years agoMerge branch 'devel' of master.kernel.org:/home/rmk/linux-2.6-mmc
Linus Torvalds [Fri, 23 Jun 2006 05:47:06 +0000 (22:47 -0700)]
Merge branch 'devel' of /home/rmk/linux-2.6-mmc

* 'devel' of master.kernel.org:/home/rmk/linux-2.6-mmc:
  [ARM] 3565/1: AT91RM9200 MMC update
  [MMC] Convert all hosts except mmci to use data->blksz

18 years agoMerge branch 'devel' of master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Fri, 23 Jun 2006 05:46:28 +0000 (22:46 -0700)]
Merge branch 'devel' of /home/rmk/linux-2.6-arm

* 'devel' of master.kernel.org:/home/rmk/linux-2.6-arm: (21 commits)
  [ARM] 3629/1: S3C24XX: fix missing bracket in regs-dsc.h
  [ARM] 3537/1: Rework DMA-bounce locking for finer granularity
  [ARM] 3601/1: i.MX/MX1 DMA error handling for signaled channels only
  [ARM] 3597/1: ixp4xx/nslu2: Board support for new LED subsystem
  [ARM] 3595/1: ixp4xx/nas100d: Board support for new LED subsystem
  [ARM] 3626/1: ARM EABI: fix syscall restarting
  [ARM] 3628/1: S3C24XX: add get_rate call to struct clk
  [ARM] 3627/1: S3C24XX: split s3c2410 clocks from core clocks
  [ARM] 3613/1: S3C2410: Add sysdev and sysclass
  [ARM] 3624/1: Report true modem control line states
  [ARM] 3620/2: ixp23xx: add uengine loader support
  [ARM] 3618/1: add defconfig for logicpd pxa270 card engine
  [ARM] 3617/1: ep93xx: fix slightly incorrect timer tick rate
  [ARM] 3616/1: fix timer handler wrap logic for a number of platforms
  [ARM] 3615/1: ixp23xx: use platform devices for physmap flash
  [ARM] 3614/1: ep93xx: use platform devices for physmap flash
  [ARM] 3621/1: fix compilation breakage for pnx4008
  [ARM] 3623/1: pnx4008: move GPIO-related defines to gpio.h
  [ARM] 3622/1: pnx4008: remove clk_use/clk_unuse
  [ARM] Enable VFP to be built when non-VFP capable CPUs are selected
  ...

18 years agoMerge branch 'devel' of master.kernel.org:/home/rmk/linux-2.6-serial
Linus Torvalds [Fri, 23 Jun 2006 05:45:53 +0000 (22:45 -0700)]
Merge branch 'devel' of /home/rmk/linux-2.6-serial

* 'devel' of master.kernel.org:/home/rmk/linux-2.6-serial:
  [ARM] 3600/1: increase amba-pl010 UART_NR to 8
  [ARM] 3571/1: netX: serial driver for Hilscher netX

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davej/cpufreq
Linus Torvalds [Fri, 23 Jun 2006 05:40:00 +0000 (22:40 -0700)]
Merge /pub/scm/linux/kernel/git/davej/cpufreq

* master.kernel.org:/pub/scm/linux/kernel/git/davej/cpufreq:
  [CPUFREQ] Fix ondemand vs suspend deadlock
  [CPUFREQ] Fix powernow-k8 SMP kernel on UP hardware bug.
  [PATCH] redirect speedstep-centrino maintainer mail to cpufreq list
  [CPUFREQ] correct powernow-k8 fid/vid masks for extended parts
  [CPUFREQ] Clarify powernow-k8 cpu_family statements

18 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Fri, 23 Jun 2006 05:15:09 +0000 (22:15 -0700)]
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6: (33 commits)
  [PATCH] myri10ge - drop workaround pci_save_state() disabling MSI
  [PATCH] myri10ge - drop workaround for the missing AER ext cap on nVidia CK804
  via-velocity: the link is not correctly detected when the device starts
  [PATCH] add b44 to maintainers
  [PATCH] WAN: ioremap() failure checks in drivers
  [PATCH] WAN: register_hdlc_device() doesn't need dev_alloc_name()
  [PATCH] skb_padto()-area fixes in 8390, wavelan
  [PATCH] make drivers/net/forcedeth.c:nv_update_pause() static
  [PATCH] network driver for Hilscher netx
  [PATCH] Dereference in tokenring/olympic.c
  [PATCH] Array overrun in drivers/net/wireless/wavelan.c
  [PATCH] Remove useless check in drivers/net/pcmcia/xirc2ps_cs.c
  [PATCH] 8139cp: add ethtool eeprom support
  [PATCH] 8139cp: fix eeprom read command length
  [PATCH] b44: update b44 Kconfig entry
  [PATCH] b44: update version to 1.01
  [PATCH] b44: add wol for old nic
  [PATCH] b44: add parameter
  [PATCH] b44: add wol
  [PATCH] b44: fix manual speed/duplex/autoneg settings
  ...

18 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Fri, 23 Jun 2006 05:11:30 +0000 (22:11 -0700)]
Merge git://git./linux/kernel/git/paulus/powerpc

* git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc: (139 commits)
  [POWERPC] re-enable OProfile for iSeries, using timer interrupt
  [POWERPC] support ibm,extended-*-frequency properties
  [POWERPC] Extra sanity check in EEH code
  [POWERPC] Dont look for class-code in pci children
  [POWERPC] Fix mdelay badness on shared processor partitions
  [POWERPC] disable floating point exceptions for init
  [POWERPC] Unify ppc syscall tables
  [POWERPC] mpic: add support for serial mode interrupts
  [POWERPC] pseries: Print PCI slot location code on failure
  [POWERPC] spufs: one more fix for 64k pages
  [POWERPC] spufs: fail spu_create with invalid flags
  [POWERPC] spufs: clear class2 interrupt status before wakeup
  [POWERPC] spufs: fix Makefile for "make clean"
  [POWERPC] spufs: remove stop_code from struct spu
  [POWERPC] spufs: fix spu irq affinity setting
  [POWERPC] spufs: further abstract priv1 register access
  [POWERPC] spufs: split the Cell BE support into generic and platform dependant parts
  [POWERPC] spufs: dont try to access SPE channel 1 count
  [POWERPC] spufs: use kzalloc in create_spu
  [POWERPC] spufs: fix initial state of wbox file
  ...

Manually resolved conflicts in:
drivers/net/phy/Makefile
include/asm-powerpc/spu.h

18 years ago[PATCH] myri10ge - drop workaround pci_save_state() disabling MSI
Brice Goglin [Fri, 23 Jun 2006 01:12:36 +0000 (21:12 -0400)]
[PATCH] myri10ge - drop workaround pci_save_state() disabling MSI

We don't need to restore the state right after saving it for later recovery
since commit 99dc804d9bcc2c53f4c20c291bf4e185312a1a0c (PCI: disable msi mode
in pci_disable_device) now prevents pci_save_state() from disabling MSI.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>